Spaces:
Sleeping
Sleeping
File size: 17,862 Bytes
c52261f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 | import os
import scipy
import torch
import random
import numpy as np
import pandas as pd
from typing import Callable
from numpy.typing import ArrayLike
from sklearn.exceptions import NotFittedError
from sklearn.preprocessing import MinMaxScaler
from src.models.nam.data import NAMDataset
from src.models.nam.trainer import Trainer
from src.models.nam.models import get_num_units
from src.models.nam.models import NAM, MultiTaskNAM
from src.models.nam.models.saver import Checkpointer
from src.models.nam.trainer.losses import make_penalized_loss_func
class NAMBase:
def __init__(
self,
units_multiplier: int = 2,
num_basis_functions: int = 64,
hidden_sizes: list = [64, 32],
dropout: float = 0.1,
feature_dropout: float = 0.05,
batch_size: int = 1024,
num_workers: int = 0,
num_epochs: int = 1000,
log_dir: str = None,
val_split: float = 0.15,
device: str = 'cpu',
lr: float = 0.02082,
decay_rate: float = 0.0,
output_reg: float = 0.2078,
l2_reg: float = 0.0,
save_model_frequency: int = 10,
patience: int = 60,
monitor_loss: bool = True,
early_stop_mode: str = 'min',
loss_func: Callable = None,
metric: str = None,
num_learners: int = 1,
n_jobs: int = None,
warm_start: bool = False,
random_state: int = 42
) -> None:
self.units_multiplier = units_multiplier
self.num_basis_functions = num_basis_functions
self.hidden_sizes = hidden_sizes
self.dropout = dropout
self.feature_dropout = feature_dropout
self.batch_size = batch_size
self.num_workers = num_workers
self.num_epochs = num_epochs
self.log_dir = log_dir
self.val_split = val_split
self.device = device
self.lr = lr
self.decay_rate = decay_rate
self.output_reg = output_reg
self.l2_reg = l2_reg
self.save_model_frequency = save_model_frequency
self.patience = patience
self.monitor_loss = monitor_loss
self.early_stop_mode = early_stop_mode
self.loss_func = loss_func
self.metric = metric
self.num_learners = num_learners
self.n_jobs = n_jobs
self.warm_start = warm_start
self.random_state = random_state
self._best_checkpoint_suffix = 'best'
self._fitted = False
def _set_random_state(self):
random.seed(self.random_state)
np.random.seed(self.random_state)
torch.manual_seed(self.random_state)
return
def _initialize_models(self, X, y):
self.num_tasks = y.shape[1] if len(y.shape) > 1 else 1
self.num_inputs = X.shape[1]
self.models = []
for _ in range(self.num_learners):
model = NAM(num_inputs=self.num_inputs,
num_units=get_num_units(self.units_multiplier, self.num_basis_functions, X),
dropout=self.dropout,
feature_dropout=self.feature_dropout,
hidden_sizes=self.hidden_sizes)
self.models.append(model)
return
def _models_to_device(self, device):
for model in self.models:
model.to(device)
return
def fit(self, X, y, w=None):
if isinstance(X, pd.DataFrame):
X = X.to_numpy()
if isinstance(y, (pd.DataFrame, pd.Series)):
y = y.to_numpy()
if isinstance(w, (pd.DataFrame, pd.Series)):
w = w.to_numpy()
self._set_random_state()
if not self.warm_start or not self._fitted:
self._initialize_models(X, y)
self.partial_fit(X, y)
return self
def partial_fit(self, X, y, w=None) -> None:
self._models_to_device(self.device)
# self._preprocessor = MinMaxScaler(feature_range = (-1, 1))
# dataset = NAMDataset(self._preprocessor.fit_transform(X), y, w)
dataset = NAMDataset(X, y, w)
self.criterion = make_penalized_loss_func(self.loss_func,
self.regression, self.output_reg, self.l2_reg)
self.trainer = Trainer(
models=self.models,
dataset=dataset,
metric=self.metric,
batch_size=self.batch_size,
num_workers=self.num_workers,
num_epochs=self.num_epochs,
log_dir=self.log_dir,
val_split=self.val_split,
test_split=None,
device=self.device,
lr=self.lr,
decay_rate=self.decay_rate,
save_model_frequency=self.save_model_frequency,
patience=self.patience,
monitor_loss=self.monitor_loss,
early_stop_mode=self.early_stop_mode,
criterion=self.criterion,
regression=self.regression,
num_learners=self.num_learners,
n_jobs=self.n_jobs,
random_state=self.random_state
)
self.trainer.train_ensemble()
self.trainer.close()
# Move models to cpu so predictions can be made on cpu data
self._models_to_device('cpu')
self._fitted = True
return self
def predict(self, X) -> ArrayLike:
if not self._fitted:
raise NotFittedError('''This NAM instance is not fitted yet. Call \'fit\'
with appropriate arguments before using this method.''')
if isinstance(X, pd.DataFrame):
X = X.to_numpy()
# X = self._preprocessor.transform(X)
X = torch.tensor(X, requires_grad=False, dtype=torch.float)
predictions = np.zeros((X.shape[0], self.num_tasks))
for model in self.models:
preds, _ = model.forward(X)
predictions += preds.detach().cpu().numpy()
# predictions = self._preprocessor.inverse_transform(predictions)
return predictions / self.num_learners
def plot(self, feature_index) -> None:
num_samples = 1000
X = np.zeros((num_samples, self.num_inputs))
X[:, feature_index] = np.linspace(-1.0, 1.0, num_samples)
feature_outputs = []
for model in self.models:
# (examples, tasks, features)
_, fnns_out = model.forward(torch.tensor(X, dtype=torch.float32))
if self.num_tasks == 1:
fnns_out = fnns_out.unsqueeze(dim=1)
# (examples, tasks)
feature_outputs.append(fnns_out[:, :, feature_index].detach().cpu().numpy())
# (learners, examples, tasks)
feature_outputs = np.stack(feature_outputs, axis=0)
# (examples, tasks)
y = np.mean(feature_outputs, axis=0).squeeze()
conf_int = np.std(feature_outputs, axis=0).squeeze()
# TODO: Scale conf_int according to units of y
# X = self._preprocessor.inverse_transform(X)
return {'x': X[:, feature_index], 'y': y, 'conf_int': conf_int}
def load_checkpoints(self, checkpoint_dir):
self.models = []
for i in range(self.num_learners):
checkpointer = Checkpointer(os.path.join(checkpoint_dir, str(i)))
model = checkpointer.load(self._best_checkpoint_suffix)
model.eval()
self.num_tasks = 1 if isinstance(model, NAM) else model.num_tasks
self.models.append(model)
self._fitted = True
return
class NAMClassifier(NAMBase):
def __init__(
self,
units_multiplier: int = 2,
num_basis_functions: int = 64,
hidden_sizes: list = [64, 32],
dropout: float = 0.1,
feature_dropout: float = 0.05,
batch_size: int = 1024,
num_workers: int = 0,
num_epochs: int = 1000,
log_dir: str = None,
val_split: float = 0.15,
device: str = 'cpu',
lr: float = 0.02082,
decay_rate: float = 0.0,
output_reg: float = 0.2078,
l2_reg: float = 0.0,
save_model_frequency: int = 10,
patience: int = 60,
monitor_loss: bool = True,
early_stop_mode: str = 'min',
loss_func: Callable = None,
metric: str = None,
num_learners: int = 1,
n_jobs: int = None,
warm_start: bool = False,
random_state: int = 42
) -> None:
super(NAMClassifier, self).__init__(
units_multiplier=units_multiplier,
num_basis_functions=num_basis_functions,
hidden_sizes=hidden_sizes,
dropout=dropout,
feature_dropout=feature_dropout,
batch_size=batch_size,
num_workers=num_workers,
num_epochs=num_epochs,
log_dir=log_dir,
val_split=val_split,
device=device,
lr=lr,
decay_rate=decay_rate,
output_reg=output_reg,
l2_reg=l2_reg,
save_model_frequency=save_model_frequency,
patience=patience,
monitor_loss=monitor_loss,
early_stop_mode=early_stop_mode,
loss_func=loss_func,
metric=metric,
num_learners=num_learners,
n_jobs=n_jobs,
warm_start=warm_start,
random_state=random_state
)
self.regression = False
def fit(self, X, y, w=None):
if isinstance(X, pd.DataFrame):
X = X.to_numpy()
if isinstance(y, (pd.DataFrame, pd.Series)):
y = y.to_numpy()
if isinstance(w, (pd.DataFrame, pd.Series)):
w = w.to_numpy()
if len(np.unique(y[~np.isnan(y)])) > 2:
raise ValueError(
'More than two unique y-values detected. Multiclass classification not currently supported.')
return super().fit(X, y, w)
def predict_proba(self, X) -> ArrayLike:
out = scipy.special.expit(super().predict(X))
return out
def predict(self, X) -> ArrayLike:
return self.predict_proba(X).round()
class NAMRegressor(NAMBase):
def __init__(
self,
units_multiplier: int = 2,
num_basis_functions: int = 64,
hidden_sizes: list = [64, 32],
dropout: float = 0.1,
feature_dropout: float = 0.05,
batch_size: int = 1024,
num_workers: int = 0,
num_epochs: int = 1000,
log_dir: str = None,
val_split: float = 0.15,
device: str = 'cpu',
lr: float = 0.02082,
decay_rate: float = 0.0,
output_reg: float = 0.2078,
l2_reg: float = 0.0,
save_model_frequency: int = 10,
patience: int = 60,
monitor_loss: bool = True,
early_stop_mode: str = 'min',
loss_func: Callable = None,
metric: str = None,
num_learners: int = 1,
n_jobs: int = None,
warm_start: bool = False,
random_state: int = 42
) -> None:
super(NAMRegressor, self).__init__(
units_multiplier=units_multiplier,
num_basis_functions=num_basis_functions,
hidden_sizes=hidden_sizes,
dropout=dropout,
feature_dropout=feature_dropout,
batch_size=batch_size,
num_workers=num_workers,
num_epochs=num_epochs,
log_dir=log_dir,
val_split=val_split,
device=device,
lr=lr,
decay_rate=decay_rate,
output_reg=output_reg,
l2_reg=l2_reg,
save_model_frequency=save_model_frequency,
patience=patience,
monitor_loss=monitor_loss,
early_stop_mode=early_stop_mode,
loss_func=loss_func,
metric=metric,
num_learners=num_learners,
n_jobs=n_jobs,
warm_start=warm_start,
random_state=random_state
)
self.regression = True
class MultiTaskNAMClassifier(NAMClassifier):
def __init__(
self,
units_multiplier: int = 2,
num_basis_functions: int = 64,
hidden_sizes: list = [64, 32],
num_subnets: int = 2,
dropout: float = 0.1,
feature_dropout: float = 0.05,
batch_size: int = 1024,
num_workers: int = 0,
num_epochs: int = 1000,
log_dir: str = None,
val_split: float = 0.15,
device: str = 'cpu',
lr: float = 0.02082,
decay_rate: float = 0.0,
output_reg: float = 0.2078,
l2_reg: float = 0.0,
save_model_frequency: int = 10,
patience: int = 60,
monitor_loss: bool = True,
early_stop_mode: str = 'min',
loss_func: Callable = None,
metric: str = None,
num_learners: int = 1,
n_jobs: int = None,
warm_start: bool = False,
random_state: int = 42
) -> None:
super(MultiTaskNAMClassifier, self).__init__(
units_multiplier=units_multiplier,
num_basis_functions=num_basis_functions,
hidden_sizes=hidden_sizes,
dropout=dropout,
feature_dropout=feature_dropout,
batch_size=batch_size,
num_workers=num_workers,
num_epochs=num_epochs,
log_dir=log_dir,
val_split=val_split,
device=device,
lr=lr,
decay_rate=decay_rate,
output_reg=output_reg,
l2_reg=l2_reg,
save_model_frequency=save_model_frequency,
patience=patience,
monitor_loss=monitor_loss,
early_stop_mode=early_stop_mode,
loss_func=loss_func,
metric=metric,
num_learners=num_learners,
n_jobs=n_jobs,
warm_start=warm_start,
random_state=random_state
)
self.num_subnets = num_subnets
def _initialize_models(self, X, y):
self.num_inputs = X.shape[1]
self.num_tasks = y.shape[1] if len(y.shape) > 1 else 1
self.models = []
for _ in range(self.num_learners):
model = MultiTaskNAM(num_inputs=X.shape[1],
num_units=get_num_units(self.units_multiplier, self.num_basis_functions, X),
num_subnets=self.num_subnets,
num_tasks=y.shape[1],
dropout=self.dropout,
feature_dropout=self.feature_dropout,
hidden_sizes=self.hidden_sizes)
model.to(self.device)
self.models.append(model)
class MultiTaskNAMRegressor(NAMRegressor):
def __init__(
self,
units_multiplier: int = 2,
num_basis_functions: int = 64,
hidden_sizes: list = [64, 32],
num_subnets: int = 2,
dropout: float = 0.1,
feature_dropout: float = 0.05,
batch_size: int = 1024,
num_workers: int = 0,
num_epochs: int = 1000,
log_dir: str = None,
val_split: float = 0.15,
device: str = 'cpu',
lr: float = 0.02082,
decay_rate: float = 0.995,
output_reg: float = 0.2078,
l2_reg: float = 0.0,
save_model_frequency: int = 10,
patience: int = 60,
monitor_loss: bool = True,
early_stop_mode: str = 'min',
loss_func: Callable = None,
metric: str = None,
num_learners: int = 1,
n_jobs: int = None,
warm_start: bool = False,
random_state: int = 42
) -> None:
super(MultiTaskNAMRegressor, self).__init__(
units_multiplier=units_multiplier,
num_basis_functions=num_basis_functions,
hidden_sizes=hidden_sizes,
dropout=dropout,
feature_dropout=feature_dropout,
batch_size=batch_size,
num_workers=num_workers,
num_epochs=num_epochs,
log_dir=log_dir,
val_split=val_split,
device=device,
lr=lr,
decay_rate=decay_rate,
output_reg=output_reg,
l2_reg=l2_reg,
save_model_frequency=save_model_frequency,
patience=patience,
monitor_loss=monitor_loss,
early_stop_mode=early_stop_mode,
loss_func=loss_func,
metric=metric,
num_learners=num_learners,
n_jobs=n_jobs,
warm_start=warm_start,
random_state=random_state
)
self.num_subnets = num_subnets
def _initialize_models(self, X, y):
self.num_inputs = X.shape[1]
self.num_tasks = y.shape[1] if len(y.shape) > 1 else 1
self.models = []
for _ in range(self.num_learners):
model = MultiTaskNAM(num_inputs=X.shape[1],
num_units=get_num_units(self.units_multiplier, self.num_basis_functions, X),
num_subnets=self.num_subnets,
num_tasks=y.shape[1],
dropout=self.dropout,
feature_dropout=self.feature_dropout,
hidden_sizes=self.hidden_sizes)
model.to(self.device)
self.models.append(model)
|